Addition below...

On Tue, Oct 26, 2004 at 09:00:50AM -0300, Fabio Olive Leite wrote:
> On Tue, Oct 26, 2004 at 01:18:44AM -0500, Jared Yanovich wrote:
> >
> > No one has tried or even thought about this? Even ideas as to how the
> > parsing may be approached are welcomed, or other constraints, etc.
>
> I have thought a bit about it, but I currently don't have any spare
> time to code anything, so I have chosen to keep quiet and "shut up
> while I can't hack".
>
> But now that comments were asked again: I'm thinking about splitting
> the rest of the line after the ---, +++ or *** (?) markers into
> substrings on the white-space characters, having the pieces after the
> first with their own white-space prepended (this is confusing, will
> try again), so that we can loop trying to find a filename with the
> first piece, and if it can't be found then we append the second piece
> and try again, and so on.
>
> Example:
>
> +++ sys/kern/Oh my god who commited spaces??? SomeTime SomeRev
>
> We get the pieces as "sys/kern/Oh", " my", " god", " who",
> " commited", " spaces???", " SomeTime" and " SomeRev". In case
> it's not a CVS diff we don't get the " SomeRev" piece.
>
> Now we loop trying to find a file named "sys/kern/Oh", and if it can't
> be found, we try "sys/kern/Oh my", and then "sys/kern/Oh my god" and
> so on.
>
> If the damn file exists, it's bound to be found by this construct. Of
> course there will be flaws in the oddest cases, but this heuristic is
> about as close to what a human would try as possible.
>
> Comments? Sorry, as I said, no patch for now...

In the cases where patches have to create non-existing files, we just
toss out the last piece if it starts with "r?[0-9]+\.[0-9]+" and then
toss out the last 5 pieces again (timestamp, hopefuly). This sucks,
but what else could be done?

--
I drowned in the universal pool of entropy
Eris has saved me, and she has set me free